A language model is a system trained on statistical and mathematical principles to predict the most likely next word, or sequence of words, given the text that came before it. That core prediction task, repeated one token at a time, is what underlies everything from autocomplete suggestions to fully generated paragraphs of text.
Modern large language models scale this idea up dramatically, training on enormous text corpora with billions of parameters, which lets them capture far more nuanced patterns in grammar, meaning, and style than earlier, smaller statistical models could. Despite the sophistication of their output, language models don't reason about truth directly, they estimate which words are statistically likely to follow, which is why they can produce fluent, confident-sounding text that's occasionally factually wrong, a limitation commonly known as hallucination.
